05 · Investigation

Technical Issue and Root-Cause Analysis

Technical investigation using deterministic comparison first, with AI helping interpret evidence before human-confirmed closure.

Proposed extension stack PythonXML Compare Utility Enterprise LLMSharePoint

Case scope: The enterprise process, systems and control foundation reflects hands-on delivery patterns. AI and agent components shown below are proposed design extensions.

Evidence comparisonIssue ownership Validated findingsResolution history

At a glance

Business problemIssues often span XMLs, logs, tickets and multiple systems, making it difficult to separate symptoms from root cause and coordinate ownership

My contribution Issue triage · comparison criteria · root-cause investigation · validation

Current environmentXML Compare Utility · test evidence · issue logs · SharePoint

Business valueFindings stay linked to evidence, owners and validated resolutions that can be reused when issues recur.

Proposed AI extension · AI / Agent / Human

Connects XML comparison results with supporting evidence to group related symptoms and narrow likely causes.

Agent

Organizes investigations, assigns findings to owners and carries validated resolutions forward when issues recur.

Human

Technical owners confirm or override hypotheses. Final root cause and resolution remain human decisions, with reasons logged.

Current workflow

01Issue intake

02Evidence collection

03Comparison

04Pattern analysis

05Root-cause validation

06Closure

What I owned

Investigation requirementsComparison criteria Severity classification criteriaEvidence requirements Issue categorizationEscalation rules Ownership modelClosure criteria Acceptance criteriaUAT AI use-case definitionAI output requirements

Proposed AI-enabled architecture

XML / logs / evidence→ XML Compare Utility + Python→ Enterprise LLM→ Investigation queue→ Owner review→ Human root-cause confirmation→ Validated resolution logged

Technical design

MVP scenarios

Scenario 1Known structural difference. XML comparison detects a known structural change. Handled by the XML Compare Utility without AI involvement.Deterministic

Scenario 2Known invalid value. A defined rule or expected value is violated. Fails deterministic check.Fail

Scenario 3Related symptoms. Multiple differences appear to stem from the same underlying issue. AI groups them and explains the relationship.Review

Scenario 4Multiple hypotheses. Evidence supports more than one possible cause. AI presents likely hypotheses and identifies what additional evidence would help.Investigation

Scenario 5Incorrect AI hypothesis. Technical owner rejects the hypothesis and records the correct interpretation.Override

Scenario 6Recurring issue. A similar failure appears later. The workflow links it to a prior validated investigation and surfaces the historical resolution.Historical match

Scenario 7Insufficient evidence. Required logs, XMLs or reproduction steps are missing. AI does not guess. The agent requests the missing evidence before analysis continues.Evidence required

Controls and governance

Deterministic comparison remains authoritative for known rules and structural differences.

AI does not declare the final root cause by itself.

Severity classification follows predefined business and technical criteria.

Missing evidence stops the investigation from progressing.

Human confirmation is required for root cause and closure.

Overrides and reasons are logged.

Only validated historical resolutions feed future investigations.

Implementation note

Azure OpenAI is the proposed LLM for grouping symptoms and generating root-cause hypotheses from validated evidence.